Lisa Piazza

Endoplasmic reticulum (ER) oxidoreductases are promising targets for host-targeted antiviral agents (HTAs) because of their essential role in protein folding, including the maturation of viral proteins. Human protein disulfide isomerase family member A3 (PDIA3) is involved in the folding of N-glycosylated client proteins, including viral envelope proteins, making it an attractive but underexplored antiviral target. Here, we describe the discovery of a novel N-hydroxyindole-2-carboxamide class of PDIA3 inhibitors, exemplified by compound AV22 (1o), which effectively inhibits PDIA3 activity and exerts broad-spectrum antiviral activity against major human respiratory viruses. Mechanistic studies indicated that AV22 (1o) increased the release of non-infectious coronavirus particles and reduced disulfide bond formation in the coronavirus spike protein, indicating that its antiviral effect stems from interference with PDIA3 activity. This new class of HTAs provides promising leads for drug discovery targeting viral envelope protein maturation vulnerabilities and is suitable for developing new BSAs.

Lipophilicity, commonly described by the n-octanol/water partition coefficient (LogP), is a key physicochemical property influencing the pharmacokinetic behavior of small molecules. Reliable LogP estimation during the early stages of drug discovery is essential to support molecular design and prioritize compounds with favorable ADMET properties. In this work, we report the development of LogPpred, an AI-based predictor of molecular lipophilicity. Starting from a curated dataset of 13,536 molecules with experimentally determined LogP values, multiple machine learning algorithms and molecular representations were systematically evaluated. The best-performing model, based on Gaussian Process Regression and RDKit molecular descriptors, achieved a mean absolute error (MAE) of 0.34 on the independent internal test set. External validation on a fully independent OECD-derived dataset yielded an MAE of 0.84, demonstrating good generalization capability across diverse chemical space. Furthermore, experimental LogP determination of an additional set of independently selected compounds confirmed the predictive reliability of the model, yielding an MAE of 0.66. Comparative analyses showed that LogPpred outperformed several widely used LogP prediction tools. Applicability domain analysis further supported the reliability of the model, with MAE values improving to 0.28 and 0.62 for in-domain compounds in the internal and external validation sets, respectively. Overall, LogPpred represents a robust, accurate, and transparent tool for the early assessment of molecular lipophilicity in medicinal chemistry and drug discovery.

Cyclin-dependent kinase 9 (CDK9) is a key regulator of transcriptional elongation and DNA repair, supporting cancer cell survival by sustaining the expression of oncogenes and anti-apoptotic proteins. Its overexpression in multiple malignancies makes it an attractive target for anticancer therapy. Here, we report a machine learning (ML) based approach to identify novel CDK9 inhibitors. Through systematic data collection and preprocessing, seventy predictive models were developed using five algorithms, two classification settings, and seven molecular representations. The best-performing model was employed to guide a virtual screening (VS) campaign, resulting in the identification of 14 compounds promising for their potential inhibitory effect. Upon enzymatic assays, two molecules with inhibitory activity in the low micromolar range were selected as promising candidates and further tested in three cancer cell lines with distinct genetic backgrounds. These experiments led to the identification of a novel compound exhibiting interesting therapeutic potential, both as a single agent and in combination with Camptothecin (CPT), revealing varying response profiles across the tested cell lines. These results illustrate the power of integrating ML within anticancer drug discovery pipelines and represent a valuable starting point for the development of novel CDK9 inhibitors.

Developing effective computational representations of protein sequences is crucial for advancing diverse areas of computational biology and bioinformatics. Ideal representations must be computationally efficient, scalable, informative, flexible across contexts, and broadly applicable. To address these requirements, we propose Protein Sequence Extended-Connectivity Fingerprints (ProSECFPs), a novel fingerprinting method inspired by Extended-Connectivity Fingerprints (ECFPs), commonly used in chemoinformatics to represent small molecules. ProSECFPs effectively capture the complex physicochemical characteristics, sequence-specific details, and structural attributes intrinsic to protein sequences. We demonstrate the effectiveness and versatility of ProSECFPs by evaluating their performance in predicting the pathogenicity of missense mutations by applying a diverse set of machine learning (ML) and deep learning (DL) algorithms. Notably, our results indicate that ProSECFPs, especially their frequency-aware variants, achieve competitive or superior accuracy compared with established protein sequence descriptors. This enhanced performance arises from their ability to comprehensively integrate amino acid composition and detailed sequence information. ProSECFPs thus provide a robust, adaptable, and highly informative computational representation of proteins, serving as a powerful foundation for addressing interdisciplinary challenges in bioinformatics, genomics, and protein engineering.

The Applicability Domain (AD) is a critical element in ensuring the reliability of in silico models for chemical safety assessment, as it defines the space within which model predictions can be considered valid. Despite its relevance, AD is often implemented using generic strategies that fail to account for the specific characteristics of the model and data set, limiting both scientific robustness and regulatory confidence. To address this limitation, we conducted a systematic benchmarking of established AD methods across regression models trained on OECD-compliant data sets, which demonstrated robust predictive ability outperforming established QSAR methods such as read-across. A regulatory-accepted approach demonstrated some limitations; additionally, its rigid structure reliant on fixed parameters and proprietary libraries limited its adaptability and transparency. We therefore reimplemented and optimized it using open-source tools to enhance flexibility, reproducibility, and predictive accuracy. Most notably, our method was demonstrated to be the preferred choice for the highest number of end points. Nonetheless, our analysis highlighted that no single strategy performs best across all scenarios. Driven by these findings and the recognized need for model-adaptive AD assessment, we developed ADvisor, a modular and broadly applicable tool that allows users to evaluate, compare, and implement AD strategies suited to the characteristics of their specific models and data sets, thereby promoting transparency and regulatory compliance in in silico applications.

Monoacylglycerol lipase (MAGL) is a key serine hydrolase involved in lipid metabolism, catalyzing the hydrolysis of monoacylglycerols into free fatty acids and glycerol. MAGL plays a central role in regulating endocannabinoid signaling and lipid homeostasis, processes often dysregulated in cancer and other pathological conditions. In recent years, MAGL has emerged as a promising therapeutic target, particularly in oncology, where its inhibition has shown potential to impair tumor growth, metastasis, and inflammation-driven processes. Alongside the development of selective MAGL inhibitors, several biochemical methods have been established to measure MAGL enzymatic activity, providing essential tools for target validation and inhibitor characterization. In this review, we provide a comprehensive and critical overview of the main approaches developed for MAGL activity evaluation, including radiometric, chromatographic, colorimetric, fluorescence-based, bioluminescence-based, and activity-based protein profiling (ABPP) assays. For each method, we discuss principles, advantages, and limitations. This review aims to support researchers in the selection of the most appropriate assay strategy for their experimental needs, ultimately fostering the rapid and accurate development of novel MAGL inhibitors with potential applications in cancer therapy and metabolic disease management.

Background/Objectives: Human carbonic anhydrases (hCAs) are metalloenzymes involved in essential physiological processes, and their selective inhibition holds therapeutic potential across a wide range of disorders. However, the high degree of structural similarity among isoforms poses a significant challenge for the design of selective inhibitors. In this work, we present a machine learning (ML)-based platform for the isoform-specific prediction and profiling of small molecules targeting hCA I, II, IX, and XII. Methods: By integrating four molecular representations with four ML algorithms, we built 64 classification models, each extensively optimized and validated. The best-performing models for each isoform were applied in a virtual screening campaign for ~2 million compounds. Results: Following a multi-step refinement process, 12 candidates were identified, purchased, and experimentally tested. Several compounds showed potent inhibitory activity in the nanomolar to submicromolar range, with selectivity profiles across the isoforms. To gain mechanistic insights, SHAP-based feature importance analysis and molecular docking supported by molecular dynamics simulations were employed, highlighting the structural determinants of the predicted activity. Conclusions: This study demonstrates the effectiveness of integrating ML, cheminformatics, and experimental validation to accelerate the discovery of selective carbonic anhydrase inhibitors and provides a generalizable framework for activity profiling across enzyme isoforms.

Rhodopsin, a G-protein-coupled receptor (GPCR) comprising the protein opsin covalently linked to the chromophore 11-cis retinal, is pivotal in visual phototransduction. Mutations in the gene encoding rhodopsin (RHO) can cause opsin misfolding or reduce its stability, resulting in retinal degenerative disorders such as retinitis pigmentosa (RP). Current therapeutic strategies employing retinoid-based chaperones partially rescue the folding and trafficking of mutant rhodopsin, but are limited by inherent toxicity and instability due to photoinduced isomerization. In the present work, a pharmacophore-based virtual screening protocol combined with molecular docking and molecular dynamics simulations was employed, leading to the identification of a novel non-retinoid opsin ligand that can potentially act as a pharmacological chaperone. Biological validation confirmed that the compound VS1 binds opsin effectively, representing a valuable starting point for structure-based optimization studies aimed at identifying new opsin stabilizers.

Protein kinases are key regulators of cellular processes and critical therapeutic targets in diseases like cancer, making them a focal point for drug discovery efforts. In this context, we developed KinasePred, a robust computational workflow that combines machine learning and explainable artificial intelligence to predict the kinase activity of small molecules while providing detailed insights into the structural features driving ligand-target interactions. Our kinase-family predictive tool demonstrated significant performance, validated through virtual screening, where it successfully identified six kinase inhibitors. Target-focused operational models were subsequently developed to refine target-specific predictions, enabling the identification of molecular determinants of kinase selectivity. This integrated framework not only accelerates the screening and identification of kinase-targeting compounds but also supports broader applications in target identification, polypharmacology studies, and off-target effect analysis, providing a versatile tool for streamlining the drug discovery process.

The application of artificial intelligence and machine learning (ML) methods is becoming increasingly popular in computational toxicology and drug design; it is considered as a promising solution for assessing the safety profile of compounds, particularly in lead optimization and ADMET studies, and to meet the principles of the 3Rs, which calls for the replacement, reduction, and refinement of animal testing. In this context, we herein present the development of VenomPred 2.0 (http://www.mmvsl.it/wp/venompred2/), the new and improved version of our free of charge web tool for toxicological predictions, which now represents a powerful web-based platform for multifaceted and human-interpretable in silico toxicity profiling of chemicals. VenomPred 2.0 presents an extended set of toxicity endpoints (androgenicity, skin irritation, eye irritation, and acute oral toxicity, in addition to the already available carcinogenicity, mutagenicity, hepatotoxicity, and estrogenicity) that can be evaluated through an exhaustive consensus prediction strategy based on multiple ML models. Moreover, we also implemented a new utility based on the Shapley Additive exPlanations (SHAP) method that allows human interpretable toxicological profiling of small molecules, highlighting the features that strongly contribute to the toxicological predictions in order to derive structural toxicophores.